[Bug tree-optimization/56365] [4.9/5/6 Regression] Missed opportunities for smin/smax standard name patterns when compiling as C++

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56365

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
           Priority|P3                          |P2
      Known to work|                            |4.3.6
   Target Milestone|---                         |4.9.4
            Summary|Missed opportunities for    |[4.9/5/6 Regression] Missed
                   |smin/smax standard name     |opportunities for smin/smax
                   |patterns when compiling as  |standard name patterns when
                   |C++                         |compiling as C++
      Known to fail|                            |4.7.3, 5.3.0, 6.0
           Severity|enhancement                 |normal

--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
In the following testcase we used to do better with GCC 4.3 at least, handling
test_02 and test_04 completely and test_01 and test_03 half-way.  Compared
to GCC 5 which only handles test_04 completely and test_03 half-way and
test_01 and test_02 not at all.

So this is a regression probably caused by adding maybe_canonicalize_comparison
to comparison folding, PR26899.

int test_01 (int a)
{
  if (127 <= a)
    a = 127;
  else if (a <= -128)
    a = -128;
  return a;
}
int test_02 (int a)
{
  if (127 < a)
    a = 127;
  else if (a <= -128)
    a = -128;
  return a;
}
int test_03 (int a)
{
  if (127 <= a)
    a = 127;
  else if (a < -128)
    a = -128;
  return a;
}
int test_04 (int a)
{
  if (127 < a)
    a = 127;
  else if (a < -128)
    a = -128;
  return a;
}